Ethan Mollick describes one of the challenges of working with AI: its capabilities are very jagged. AI can be really good at some tasks but terrible at others, and it's not always easy to predict where it's most useful. When weaknesses that create bottlenecks are identified, AI companies focus development in those areas, Just because something is a weakness now doesn't necessarily mean AI will never be able to do that task.
You can see how AI is indeed superhuman in some areas, but in others it is either far below human level or not overlapping at all. If this is true, then AI will create new opportunities working in complement with human beings, since we both bring different abilities to the table.
The exact abilities of AI are often a mystery, so it is no wonder AI is harder to use than it seems.
A system is only as functional as its worst components. We call these problems bottlenecks. Some bottlenecks are because the AI is stubbornly subhuman at some tasks.
Bottlenecks can create the impression that AI will never be able to do something, when, in reality, progress is held back by a single jagged weakness. When that weakness becomes a reverse salient, and AI labs suddenly fix the problem, the entire system can jump forward.